Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-13778

LNet Router: bug in routing selection algorithm

    XMLWordPrintable

Details

    • Bug
    • Resolution: Unresolved
    • Minor
    • None
    • None
    • 3
    • 9223372036854775807

    Description

      If we have previously selected a source NID to send from, we need to select a routed network which we can reach from the source NID we have specified. The current logic considers src NID after we've selected the remote net. When we look for a gateway to that remote net, since we didn't consider the src NID ahead of time we could end up not finding a gateway on the same net as the src NID, and then fail the send with host EHOSTUNREACH

      This can happen in a setup as follows:

      src NID A -> GATEWAY A -> remote Net 1
      src NID B -> GATEWAY B -> remote Net 2

      Both remote Nets 1 and 2  are reacheable via two different gateways. However, we want to restrict on src NID A. The current algorithm could give us Gateway B which would result in a EHOSTUNREACH

      Another issue here is that when source NID is specified we want to end up sending to the same destination NID. This is to ensure that we keep the original NI selection by the initiator which could be NUMA optimal.

      Attachments

        Activity

          People

            ashehata Amir Shehata (Inactive)
            ashehata Amir Shehata (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

            Dates

              Created:
              Updated: